fff59a5f4f
nvk: Fix descriptor offset alignment
...
This fixes two bugs. First, we were aligning when setting the bindign
offset but not actually applying that alignment to buffer_size so when
we incremented by stride * descriptorCount, it incremented starting with
the unaligned value. Second, in some cases we return stride = align = 0
and the ALIGN_POT macro will return 0 when something's aligned to 0.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
443717a17e
nvk: Properly indent a comment
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
98a573eb15
nvk: Implement push constants
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
717ed1c48f
nvk: Load the requested descriptor size
...
We were hard-coding to vec4s for some reason. This results in vec4
texture handles which, while properly ignroed by codegen, are just
weird and not what we want.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
b6483f5126
nvk: Handle cube storage images properly
...
TIC entries for cubes divide the array size by 6 so we don't want that
when using it as a storage image. This means nvk_image_view needs
separate TIC entries for storage vs. sampled now.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
8b08bbe2cf
nvk: Use a descriptor type instead of a hand-rolled thing
...
This was an attempt to prematurely optimize some things. Sure, it's a
hot-path, but there's other control-flow in there and we can optimize it
later if needed.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
0648f0fa03
nvk: Advertise non-zero descriptor set limits
...
Right now, everything's pretty much unlimited. We may want to limit
UBOs in the future to just what we can bind but everything else will
probably remain pretty high.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
aa78f098f2
nvk: Advertise descriptor array indexing
...
This isn't as dynamic as it looks. It still requires the index to be
uniform, it just lets it be not a constant. Given how our descriptor
sets work, this is all handled trivially already.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
dda149c053
nvk: Better advertise image format features
...
We now advertise the top-level features and we don't falsely claim
write-without-format support per-format. In theory, NV hardware should
be able to do shaderStorageImageReadWithoutFormat but codegen doesn't
know how yet.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
db81c26525
nvk: Add support for dynamic buffers
...
These are recorded in the CPU portion of the descriptor set and combined
with the dynamic offset at vkCmdBindDescriptorSets time and placed into
the root descriptor table.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
1f7476a03d
nvk: Use the correct root descriptor table size for CmdDispatch
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
fae4f1884c
nvk: Advertise KHR_dedicated_allocation
...
We're not requiring or even requesting anything to be dedicated yet but
we may as well put in the plumbing. This gets more tests running.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
fedff04fbd
nvk: Implement buffer views
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
1f72430bd6
nvk: Move is_storage_image_format to nvk_format.c
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
788e94294f
nil: Add a helper for filling out buffer TIC entries
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
834f01f1b1
nvk: Implement samplers
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
100c7060d8
nvk: Re-format nvk_sampler.c
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
cb4eff26e7
nvk: Teture pool sizes are maximums not sizes
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
e13a64d286
nil: Fix image array layer alignments
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
f3ef6f2ee1
nil: Refactor TIC image extent setup
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
6a6cf64a78
nil: Don't minify image dimensions when setting up TIC
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
a67cfe2590
nil: Drop miptail support for now
...
We don't have any idea how they work or even which TIC bits have to do
with them. Better off just leave them alone for now.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
974212a581
nouveau/classes: add 906f header support.
...
This needed some tweaks to the generator script
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
7a86192d3a
nvk: Drop vk_sync BO refs after push_submit
...
Otherwise, all syncs are cumulative and, if the same command buffer is
submitted multiple times, it will sync on the union of all syncs it's
ever seen.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
4c485b0b6f
nouveau/ws: add a push reset just for references.
...
This is to be used with the empty submission pushbuf.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
e2565c2fc2
nvk: fix empty cmd submission.
...
The sync obj stuff does a second queue submit with just signals
and no command buffers. Make sure things get sent to the kernel
so sync works.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
bf01f4d66a
nvk: Support compressed images in copy commands
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
20d8d1e239
nvk: Add a more competent GetPhysicalDeviceImageFormatProperties
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
fc19173014
nvk: Rework format features queries
...
Split the buffer and image portions into vk_buffer_view.c and vk_image.c
respectively. This way they're a lot closer to each other. Put the
main combined helper in nvk_format.c since it's a format thing. Also,
restructure things a bit based on ANV to make the code look nicer.
Also, implement VK_KHR_format_feature_flags2 while we're here.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
76253dbf94
nvk: Reset and properly clean up command buffer upload areas
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
2a914028f1
nvk: Implement VkGetImageSubresourceLyout
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
23e6689b38
nvk: Re-indent image entrypoints
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
04d2bf2ee7
nvk: Convert to using NIL for image layout
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:55 +00:00
eefb60b832
nvk: Manually offset for array layers in copy/blit
...
The more I start at the hardware the more convinced I become that arrays
and 3D are not the same thing. I'm sure they're similar enough in
certain circumstances but they don't appear to be in the way they're
treated in the 2D hardware, especially when miplevels are involved.
Instead of trying to use the 2D engine's depth/slice logic, offset
manually. We have to emit a copy command for each slice anyway so
making the GPU do those calculations doesn't really gain us anything.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
d673111bc8
Revert "nvk: Stop returning VK_ERROR_FORMAT_NOT_SUPPORTED for non-blitable"
...
This reverts commit 74ad75f2624c50f732db35946a127eb11edee3e5.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
3e706a974e
nvk: Move Fill/UpdateBuffer to nvk_cmd_copy
...
Even if they technically use blit-like HW commands, they're much more a
copy operation than a blit since there's no scaling going on.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
3beb4daa2b
nvk: Align arguments consistently in copy/blit code
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
b647527476
nvk: Use helpers for push_ref
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
3516136d7a
nvk: Add an nvk_get_format helper
...
This makes the array size internal to nvk_format.c so we don't have to
keep it in sync when we add extra elements. Also, there's no reason why
we need to be hand-rolling that loop all the time.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
ec6d56d4ce
nil: Add a nil_view and code to fill out TIC entries
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
c12c94cde3
nil: Add the TIC format table from nouveau
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
69967e3f0d
nil: Create images
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
13abd1dbf2
nouveau: Add stubs for an image layout library called NIL
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
5623de7345
nouveau/ws: initial debugging options for command submissions
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
95843dd754
nouveau/headers: Generate parser functions
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
43444e1a72
nvk/blit: assert that formats are supported
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
1de656bce3
nouveau/ws: assert on broken channel
...
no point in continuing
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
008e9768ad
nouveau/ws: make sure we don't submit nonsense
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
ee50fe7d85
nvk: Rework descriptor writes
...
In particular, this makes us only touch the descriptor UBO memory as a
memcpy destination unless we're doing a descriptor copy. This should be
much better for mapped GPU memory than the read-modify-write required by
our previous implementation of image descriptor writes with their 12:20
configuration.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
7d6ea3fcf5
nvk: Reserve a null image descriptor
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326 >
2023-08-04 21:31:54 +00:00
Faith Ekstrand